0.23.0

New Features 🎉

• Aggregate-on-JOIN support: a full pipeline for running aggregates over join results, including:
  • 3+ table join support and DISTINCT aggregates in AggregateScan.
  • LEFT/RIGHT/FULL JOIN and COUNT(DISTINCT).
  • HAVING clause and per-aggregate FILTER clause.
  • ORDER BY within STRING_AGG and ARRAY_AGG.
  • JSON sub-field GROUP BY in the DataFusion aggregate path.
  • Date/Timestamp projection and STDDEV/VARIANCE aggregates.
  • Extended aggregate TopK detection to GROUP BY column and MIN/MAX ordering.
  • TopK optimization for aggregate queries via Tantivy and DataFusion.
• New edge_ngram tokenizer for search-as-you-type use cases.
• Support for IS NULL/IS NOT NULL predicates in Top-K JoinScan.
• Per-field tunable BM25 k1 and b parameters via typmod.
• Explicit control of prefix behavior for fuzzy queries.
• keep_whitespace option for Lindera tokenizers.
• Support for the citext Postgres type.

Performance Improvements 🚀

• Lazily set up tokenizers, avoiding unnecessary initialization overhead.
• Conditionally enable scoring in row estimation for faster planning.
• Cache anyelement_search_opoids to reduce repeated catalog lookups.
• Cache score/snippet function OIDs.
• Cache segment_id to segment_ordinal mapping during SearchIndexReader construction.

Stability Improvements 💪

• Suppress LIMIT pushdown when non-pushable post-filters are present.
• Fix Top K for prepared statements.
• Recognize identity expressions (e.g. id + 0) in JoinScan ORDER BY matching.
• Re-add meaningful term filter.
• Fix boolean comparison edge cases.
• Allow unaliased, tokenized index fields to be selected by name when the same column is indexed multiple times.
• Support varchar, text[], and citext parameters in operators with generic plans.

The full changelog is available on the GitHub Release.